Arabic GSM8K is an Arabic translation of the GSM8K (Grade School Math 8K) dataset, which contains high-quality linguistically diverse grade school math word problems. The original dataset was created to support the task of question answering on basic mathematical problems that require multi-step reasoning, and this Arabic version aims to extend these capabilities to Arabic language models and applications.
The dataset maintains the same characteristics as the original GSM8K:
This dataset is designed to test mathematical reasoning capabilities in Arabic language models. It can be used to benchmark Arabic LLMs on tasks requiring logical and mathematical reasoning.
The text in the dataset is in Arabic.
Each instance contains a string for the grade-school level math question in Arabic and a string for the corresponding answer with multiple steps of reasoning and calculator annotations.
Example:
{
'question': 'باعت نتاليا مشابك إلى 48 من أصدقائها في شهر أبريل، ثم باعت نصف ذلك العدد من المشابك في شهر مايو. كم عدد المشابك التي باعتها نتاليا في شهري أبريل ومايو معًا؟',
'answer': 'باعت ناتاليا 48 ÷ 2 = <<48/2=24>>24 مشبكًا في مايو. باعت ناتاليا 48 + 24 = <<48+24=72>>72 مشبكًا في أبريل ومايو مجتمعين. #### 72',
}
question in Arabic. It contains multiple steps of reasoning with calculator annotations and the final numeric solution.The dataset follows the same split structure as the original GSM8K:
| name | train | test |
|---|---|---|
| main | 7473 | 1319 |
The Arabic GSM8K dataset was created to extend the capabilities of the original GSM8K dataset to Arabic-speaking users and Arabic language models. Mathematical reasoning is a fundamental capability for language models, and having this resource in Arabic helps advance the development of Arabic AI systems with strong mathematical reasoning abilities.
The dataset is a translation of the original GSM8K dataset. The original problems and their solutions were translated from English to Arabic using large language models with human evaluation, maintaining the mathematical integrity and reasoning steps of the original problems.
When using this dataset, researchers should be aware that it is a translation of the original GSM8K. While efforts have been made to ensure accuracy in translation, there may be linguistic nuances or cultural contexts that differ between the English and Arabic versions.
If you use this dataset in your research, please cite both the original GSM8K dataset and this Arabic adaptation:
@article{cobbe2021gsm8k,
title={Training Verifiers to Solve Math Word Problems},
author={Cobbe, Karl and Kosaraju, Vineet and Bavarian, Mohammad and Chen, Mark and Jun, Heewoo and Kaiser, Lukasz and Plappert, Matthias and Tworek, Jerry and Hilton, Jacob and Nakano, Reiichiro and Hesse, Christopher and Schulman, John},
journal={arXiv preprint arXiv:2110.14168},
year={2021}
}
@misc{arabic-gsm8k,
title={Arabic GSM8K: Arabic Grade School Math Dataset},
author={Omartificial-Intelligence-Space},
year={2025},
howpublished={\url{https://huggingface.co/datasets/Omartificial-Intelligence-Space/Arabic-gsm8k}}
}
Arabic GSM8K is an Arabic translation of the GSM8K (Grade School Math 8K) dataset, which contains high-quality linguistically diverse grade school math word problems. The original dataset was created to support the task of question answering on basic mathematical problems that require multi-step reasoning, and this Arabic version aims to extend these capabilities to Arabic language models and applications.
The dataset maintains the same characteristics as the original GSM8K:
This dataset is designed to test mathematical reasoning capabilities in Arabic language models. It can be used to benchmark Arabic LLMs on tasks requiring logical and mathematical reasoning.
The text in the dataset is in Arabic.
Each instance contains a string for the grade-school level math question in Arabic and a string for the corresponding answer with multiple steps of reasoning and calculator annotations.
Example:
{
'question': 'باعت نتاليا مشابك إلى 48 من أصدقائها في شهر أبريل، ثم باعت نصف ذلك العدد من المشابك في شهر مايو. كم عدد المشابك التي باعتها نتاليا في شهري أبريل ومايو معًا؟',
'answer': 'باعت ناتاليا 48 ÷ 2 = <<48/2=24>>24 مشبكًا في مايو. باعت ناتاليا 48 + 24 = <<48+24=72>>72 مشبكًا في أبريل ومايو مجتمعين. #### 72',
}
question in Arabic. It contains multiple steps of reasoning with calculator annotations and the final numeric solution.The dataset follows the same split structure as the original GSM8K:
| name | train | test |
|---|---|---|
| main | 7473 | 1319 |
The Arabic GSM8K dataset was created to extend the capabilities of the original GSM8K dataset to Arabic-speaking users and Arabic language models. Mathematical reasoning is a fundamental capability for language models, and having this resource in Arabic helps advance the development of Arabic AI systems with strong mathematical reasoning abilities.
The dataset is a translation of the original GSM8K dataset. The original problems and their solutions were translated from English to Arabic using large language models with human evaluation, maintaining the mathematical integrity and reasoning steps of the original problems.
When using this dataset, researchers should be aware that it is a translation of the original GSM8K. While efforts have been made to ensure accuracy in translation, there may be linguistic nuances or cultural contexts that differ between the English and Arabic versions.
If you use this dataset in your research, please cite both the original GSM8K dataset and this Arabic adaptation:
@article{cobbe2021gsm8k,
title={Training Verifiers to Solve Math Word Problems},
author={Cobbe, Karl and Kosaraju, Vineet and Bavarian, Mohammad and Chen, Mark and Jun, Heewoo and Kaiser, Lukasz and Plappert, Matthias and Tworek, Jerry and Hilton, Jacob and Nakano, Reiichiro and Hesse, Christopher and Schulman, John},
journal={arXiv preprint arXiv:2110.14168},
year={2021}
}
@misc{arabic-gsm8k,
title={Arabic GSM8K: Arabic Grade School Math Dataset},
author={Omartificial-Intelligence-Space},
year={2025},
howpublished={\url{https://huggingface.co/datasets/Omartificial-Intelligence-Space/Arabic-gsm8k}}
}